Continuous Departure Time Models: A Bayesian Approach

Models of departure time presently rely on discrete choice or simple proportions across blocks of time within a 24-hour day. Duration models allow for more flexible specifications to explain both unimodal (one-peak) and multi-modal (multi-peak) data, which are common in (aggregate) departure time data. This paper offers Bayesian estimates of continuous departure time models using accelerated failure time (AFT) specifications for various trip purposes with several distributional specifications, including the lognormal, Weibull, Weibull (with and without unobserved heterogeneity), and a mixture of normals. The home-based work (HBW) and non-home -based (NHB) trip models are modeled using unimodal distributions, while the home-based non-work (HBNW) trip departure times are modeled via a bi-modal distribution. The results indicate that a Weibull with unobserved heterogeneity performs well among unimodal distributions, and that multi-peak profile can be modeled well with a mixture of normals.